Machine Learning-Based Predictive Maintenance Scheduling in Industrial Settings

Abstract

This research examines the use of machine learning (ML) techniques on predictive maintenance scheduling within
industrial settings. The current trend in modern industries is the use of complex and highly automated systems. Often,
there is a need to maximize system uptime while minimizing maintenance costs. Inefficient, reactive, unplanned
downtimes, and excessive servicing are symptoms of inadequate advanced predictive or classically defined
maintenance strategies. ML predictive maintenance attempts to enhance scheduling by anticipating failures based on
historical and real-time data received from sensors on the equipment. This work is based on a comprehensive model
that incorporates supervised ML algorithms—namely, Condition Monitoring Systems (CMS) with Random Forest
and Gradient Boosting—to predict machine failure. The work includes data preprocessing, feature selection, model
training, validation, deployment of the model on a scheduler, and operational framework. Performance evaluation
confirms the superiority of the ML approach compared to time-based or reactive maintenance strategies, enhancing
reliability while reducing operational costs. A case study conducted in a manufacturing plant showed an improvement
in fault detection accuracy and maintenance lead-time prediction. The results accentuate the impact ML poses over
maintenance planning in industrial settings. Future work could look at integrating advance adaptive maintenance
strategies through digital twins and reinforcement learning.